Machine Learning Integration of Ultrasound Habitat Radiomics for Predicting Axillary Metastasis in Clinically Node-negative Triple-negative Breast Cancer.
Authors
Affiliations (5)
Affiliations (5)
- Department of Ultrasound, Shanghai General Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai, China.
- Department of Ultrasound, Xuzhou Central Hospital, Southeast University, Xuzhou, Jiangsu, China.
- School of Health Science and Engineering, University of Shanghai for Science and Technology, Shanghai, China.
- Department of Ultrasound, The First Affiliated Hospital of Nanjing Medical University, Nanjing, China.
- Department of Ultrasound, Shanghai General Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai, China. Electronic address: [email protected].
Abstract
To develop and validate habitat radiomics as a biomarker for predicting axillary lymph node metastasis (ALNM) in clinically node-negative (cN0) patients with triple-negative breast cancer (TNBC). A total of 313 TNBC patients from two institutions between January 2021 and December 2024 with cN0 were retrospectively included. Habitat radiomics features and conventional radiomics features were both extracted from grayscale ultrasound images. Tumor habitat heterogeneity features were derived using superpixel-based segmentation, followed by the construction of a habitat radiomics signature (HRS) and a conventional radiomics score (CRS). Dimensionality reduction and feature selection of radiomics features were conducted using the least absolute shrinkage and selection operator. Machine learning algorithms were then used to construct predictive models with 10-fold cross-validation. Model performance was evaluated using receiver operating characteristic analysis, calibration curves, and decision curve analysis. Seven radiomics features were used to construct the CRS, and nine radiomics features were used to construct the HRS. The HRS demonstrated superior diagnostic performance over the CRS for predicting ALNM (0.84 vs. 0.72 in the training and 0.82 vs. 0.69 in the validation cohorts, both p < 0.05). In the validation cohort, among the integrated machine learning models, XGboost achieved the highest diagnostic performance (AUC = 0.89, 95%CI 0.83-0.95), outperforming both the CRS and HRS (p < 0.05). Decision curve analysis showed a net benefit at clinically relevant thresholds (5%-40%) for predicting ALNM in TNBC. The SHapley Additive exPlanations analysis identified the HRS as the top contributor. The habitat radiomics-based machine learning model shows potential for non-invasively predicting ALNM in cN0 TNBC patients.